Open access Original research J Immunother Cancer: first published as 10.1136/jitc-2020-001429 on 13 November 2020. Downloaded from Radiomics to predict outcomes and abscopal response of patients with cancer treated with immunotherapy combined with radiotherapy using a validated signature of CD8 cells
1,2,3 4 5 5 Roger Sun , Nora Sundahl, Markus Hecht , Florian Putz, Andrea Lancia,6 Angela Rouyar,2 Marina Milic,2 Alexandre Carré,1,2 Enzo Battistella,2 Emilie Alvarez Andres,2,7 Stéphane Niyoteka,2 Edouard Romano,1 Guillaume Louvel,1 Jérôme Durand-Labrunie,1 Sophie Bockel,1,2,3 8 1,2 9 Rastilav Bahleda, Charlotte Robert , Celine Boutros , 10 7,10 5 9 Maria Vakalopoulou, Nikos Paragios, Benjamin Frey , Jean-Charles Soria, 3,8 9 5 4 5 Christophe Massard, Charles Ferté, Rainer Fietkau, Piet Ost, Udo Gaipl , Eric Deutsch1,2,3
To cite: Sun R, Sundahl N, ABSTRACT free survival (HR=1.67, p=0.040), out-of-field abscopal Hecht M, et al. Radiomics to Background Combining radiotherapy (RT) with immuno- response (AUC=0.70, p=0.014) and overall survival predict outcomes and abscopal oncology (IO) therapy (IORT) may enhance IO-induced (HR=2.08, p=0.023), which remained significant in a response of patients with cancer antitumor response. Quantitative imaging biomarkers can multivariate analysis including clinical and biological treated with immunotherapy variables. combined with radiotherapy be used to provide prognosis, predict tumor response Conclusions These results enhance the predictive using a validated signature in a non-invasive fashion and improve patient selection of CD8 cells. Journal for for IORT. A biologically inspired CD8 T-cells-associated value of the biologically inspired CD8 radiomics score ImmunoTherapy of Cancer radiomics signature has been developed on previous and suggests that tumor heterogeneity should be 2020;8:e001429. doi:10.1136/ cohorts. We evaluated here whether this CD8 radiomic systematically considered in patients treated with IORT. jitc-2020-001429 signature is associated with lesion response, whether This CD8 radiomics signature may help select patients it may help to assess disease spatial heterogeneity for who are most likely to benefit from IORT. Accepted 29 September 2020 predicting outcomes of patients treated with IORT. We http://jitc.bmj.com/ also evaluated differences between irradiated and non- irradiated lesions. INTRODUCTION Methods Clinical data from patients with advanced solid The introduction of immune-checkpoint tumors in six independent clinical studies of IORT were inhibitors has profoundly changed the investigated. Immunotherapy consisted of 4 different immuno-oncology (IO) and the treatment 1 2 drugs (antiprogrammed death-ligand 1 or anticytotoxic of patients with metastatic cancers. Unfor- on September 25, 2021 by guest. Protected copyright. T-lymphoc yte-associa ted protein 4 in monotherapy). Most tunately, despite the use of patients stratifica- patients received stereotactic RT to one lesion. Irradiated tion criteria, the average response rate (RR) and non-irradiated lesions were delineated from baseline remains low while being significantly volatile and the first evaluation CT scans. Radiomic features (20%–50%).3 Strong evidence indicates that were extracted from contrast-enhanced CT images and radiotherapy (RT) can invoke both local and the CD8 radiomics signature was applied. A responding systemic immunostimulatory effects that could lesion was defined by a decrease in lesion size of at least synergize with immunotherapy in systemic 30%. Dispersion metrices of the radiomics signature were tumor control.4–8 However, open questions estimated to evaluate the impact of tumor heterogeneity in about modalities to improve IORT efficacy, © Author(s) (or their patient’s response. employer(s)) 2020. Re-use such as the choice and number of lesions to Results A total of 94 patients involving multiple lesions 9 10 permitted under CC BY. (100 irradiated and 189 non-irradiated lesions) were irradiate, are still pending. Patient’s strat- Published by BMJ. considered for a statistical interpretation. Lesions with ification becomes a necessity and the devel- For numbered affiliations see high CD8 radiomics score at baseline were associated opment of innovative biomarkers to optimize end of article. with significantly higher tumor response (area under the IORT paves the way to precision medicine in Correspondence to receiving operating characteristic curve (AUC)=0.63, cancer care and improved clinical outcomes. Prof. Eric Deutsch; p=0.0020). Entropy of the radiomics scores distribution on Quantitative imaging biomarkers are eric. deutsch@gustaveroussy. fr all lesions was shown to be associated with progression- of great interest. Medical images are
Sun R, et al. J Immunother Cancer 2020;8:e001429. doi:10.1136/jitc-2020-001429 1 Open access J Immunother Cancer: first published as 10.1136/jitc-2020-001429 on 13 November 2020. Downloaded from non-invasive, part of standard clinical protocol and confounding predictors of response (PFS, OS) and of reflect the whole tumor burden for which every single abscopal response were analyzed. lesion can be analyzed, contrary to traditional biopsy.11 This is of crucial importance with IORT, where disease Image analysis and treatment encompass and have to account for intra- Definition of analyzed lesions patient heterogeneity. Lesions have different microenvi- Analyzed lesions were any tumor (primary or secondary) ronments and may receive either local and/or systemic that was identifiable on both baseline and follow-up CTs, treatment effects leading to heterogeneous patterns of with diameter ≥5 mm at baseline. Lesions that could not global response.12 be accurately discriminated from surrounding tissues Evidence-based radiomic biomarkers have shown prom- (ie, lung nodule adjacent or within atelectasis) or from ising results to predict response to immunotherapy,13–18 other adjacent lesions at baseline or follow-up CTs (ie, but to the best of our knowledge, there is no evidence confluent metastases) were not delineated and excluded. yet in IORT. Our team previously validated a biologically Irradiated and non-irradiated lesions were labeled. inspired radiomics-signature trained to estimate from conventional contrast-enhanced CT scans the abun- Feature extraction and CD8 T-cells radiomics score computation dance of CD8 T-cells and demonstrated correlations with Radiomic features were extracted from contrast- outcomes of patients treated with immunotherapy.13 enhanced CTs. All images had slice-thickness ≤5 mm and Here, we aimed to evaluate in a cohort of patients treated were reconstructed using soft or standard convolution with IORT whether this tool can predict tumor response kernels. Experienced physicians from one center (RS, at a lesion-based level. We evaluated also whether this AL) delineated lesions on baseline and follow-up scans. radiomics signature could help to assess disease spatial All the segmentations were reviewed by one physician heterogeneity at a patient-level for clinical outcome (RS) to ensure a homogeneous segmentation method in prediction, and finally evaluated the impact of irradiation accordance with the published signature. Two volumes on the lesion radiomic scores. of interest (VOIs) were segmented for each lesion: (1) the lesion itself and (2) a peripheral ring, created following the method published in the original study13 to compute the radiomics signature. The ring was created METHODS using automated dilatation and shrinkage of the tumor Data and study design boundaries by 2 mm on each side, namely outside and Patients with metastatic solid tumors from three centers inside the boundary, resulted in a ring with a thickness included in clinical trials of IORT were retrospectively of 4 mm (online supplementary figure S1). Radiomic screened for inclusion. feature extraction was performed using LIFEx software Patients for whom contrast-enhanced CT scans were V.3.44 (Local Image Feature Extraction, freeware, www. available at both baseline and after start of treatment lifexsoft.org).22 Images were resized to 1×1×1 mm3 voxels
(first follow-up) were included for analysis. using three-dimensional Lagrangian polygon interpo- http://jitc.bmj.com/ Six clinical trials were considered for this lation. Hounsfield-units (HU) values in the VOIs were study (SABR-PDL1: NCT02992912, MEL-IPI-RX: then resampled into 400 discrete values using absolute NCT01557114, ‘ghent-nivo’: NCT02821182,19 ‘ghent-ipi: discretisation.23 The minimum and maximum bounds of NCT02406183,20 ‘ghent-pembro’: NCT02826564,21 the resampling interval were set to −1000 and 3000 HU, ST-ICI: NCT03453892). Immunotherapy consisted of leading to a bin size of 10 HU. anticytotoxic T-lymphocyte- associated protein 4 (ipilim- The previously validated radiomics score of CD8 T-cells on September 25, 2021 by guest. Protected copyright. umab), antiprogrammed cell death protein 1 (anti-PD-1) consists of a linear interpolation on the basis of eight vari- (pembrolizumab, nivolumab) or antiprogrammed death- ables: five radiomic features extracted from each lesion, ligand 1 (anti-PD- L1) (atezolizumab) in monotherapy. two discrete labels about lesion location and one imaging The main RT regimen was stereotactic RT delivered in acquisition-related variable—the peak kilovoltage three fractions after the start of immunotherapy. Clinical, (kVp).13 The five features from the radiomics signature biological and imaging data at baseline and follow-up, were extracted and normalized in the range 0–1. VOI and PD-L1 expression were retrieved when available. location was labeled as adenopathy, head and neck, lung, The primary objective was to evaluate the association liver or other (subcutaneous or abdominal lesions) and between the previously validated biologically inspired the kVp was retrieved from CT metadata. Coefficients radiomics score of CD8 T-cells of a lesion and its response from the validated signature were applied to these eight at the first follow-up CT.13 Secondary objectives were variables to obtain the radiomics score of each analyzed (1) to evaluate the association between the radiomics lesion (figure 1). scores of the analyzed lesions (see Definition of analyzed lesions) and patients’ clinical outcomes at the first evalu- Response evaluation ation, progression-free survival (PFS) and overall survival A lesion-wise evaluation of relative change in diameter (OS) and (2) to assess differences between irradiated and between baseline and follow-up was carried out using non-irradiated lesions. Potential clinical and biological RECIST V.1.1 criteria. A responding lesion was defined
2 Sun R, et al. J Immunother Cancer 2020;8:e001429. doi:10.1136/jitc-2020-001429 Open access J Immunother Cancer: first published as 10.1136/jitc-2020-001429 on 13 November 2020. Downloaded from
Figure 1 Radiomics workflow. Irradiated and non-irradiated lesions were delineated. After image preprocessing to discretise the voxel intensity and to resample the voxel size, features from the radiomics signature of CD8 T-cells published in Sun et al, Lancet Oncol 2018, were extracted, allowing the estimation of CD8 T-cells. Associations with outcomes were evaluated. by a decrease in lesion size of >30%. A progressive lesion Statistical analysis was defined by an increase in lesion size of >20%. To take Comparisons between variables were performed using into account the overall tumor burden and to assess the Wilcoxon signed-rank test or Kruskal-W allis test for contin- different patterns of response on a patient basis, mixed uous variables, and Fisher’s exact test for categorical vari-
response was defined as the presence of both progressive ables. Clustering of patients into two distinct groups was http://jitc.bmj.com/ and responding lesions, as opposed to patients presenting done on the basis of the median value, unless there was a only responding (uniform response) or progressive validated cut-off (ie, biological variables). A threshold of lesions (uniform progression), irrespective of stable <0.05 was defined for double-tailed p value’s significance. lesions or only stable lesions (stable disease (SD)). To Statistical analyses were performed using R software compare patterns of response of non-irradiated lesions, V.3.6.0 (https://www.r-project.org/).25 OS and PFS were
occurrence of response in any non-irradiated lesion computed according to the Kaplan-Meier method and on September 25, 2021 by guest. Protected copyright. was defined as lesion-based abscopal response, while Cox proportional-hazards survival estimates. End points a decrease of >30% in the sum of the largest diameter were death from any cause for OS, and any recurrence or of non-irradiated lesion was defined as a RECIST-based death for PFS. Multivariate models included the clinically abscopal response. relevant variables that were significant in univariate anal- ysis. No imputation was made for the missing data. Spatial heterogeneity evaluation Role of the funding sources To assess the impact of intrapatient interlesion CD8 The funders of the study had no role in study design, data radiomics score heterogeneity, minimum value, maximal collection, data analysis, data interpretation or writing of value, mean value, SD and entropy of the lesion scores the report. ED and RS had full access to all the data in distribution were retrieved. For subgroup analysis of irra- the study and had final responsibility for the decision to diated and non-irradiated lesions, when the number of submit for publication. lesions was >1, the minimum value was kept. Median value was used to separate patients into two groups. Patients with extreme values (
Sun R, et al. J Immunother Cancer 2020;8:e001429. doi:10.1136/jitc-2020-001429 3 Open access J Immunother Cancer: first published as 10.1136/jitc-2020-001429 on 13 November 2020. Downloaded from
RESULTS responses had significantly higher OS (HR=0.20, 95% CI Patient characteristics and analyzed lesions (0.08 to 0.53), p=0.0011), while OS of patients with mixed Ninety-four patients were included from 1 July 2011 to response was not significantly different from that of 31 March 2019. Baseline characteristics of the patients patients with uniform progression (p=0.84) (figure 2A). are presented in table 1. All patients were diagnosed with advanced or metastatic solid tumors. Median age at base- Out-of-field response line was 60.38 (IQR 50.68–69.04). For the 91 patients with non-irradiated lesions avail- Two patients (2.1%) with irradiated brain metastases able for analysis, out-of- field response using aggregate had no contrast-enhanced brain CTs available for imaging diameter of non-irradiated lesions according to RECIST analysis of the irradiated lesion but were kept for the (RECIST-based abscopal response) at the first follow-up analysis of non-irradiated lesions response. Conversely, CT was 16.5% (n=15). Using response defined by 30% three patients (3.2%) had no identifiable non-irradiated reduction in any single non-irradiated lesions (lesion- analyzed lesion but were considered for the irradiated based abscopal response), the non-irradiated RR was lesion response. Overall, 100 irradiated lesions and 189 26.4% (n=24). While both definitions were significantly non-irradiated lesions were analyzed at baseline and associated with OS (p=0.011 and p=0.041, respectively), follow-up, leading to a total of 578 lesions. At baseline, only patients with RECIST-based abscopal response had median size of irradiated lesions was larger than non- significantly higher OS than those with no out-of-field irradiated lesions (26.9 mm, IQR (17.6–37.7) vs 21.5 mm, response (HR=0.31, 95% CI (0.11 to 0.88), p=0.027), IQR (15.2–31.9), respectively, difference in location=3.6, contrary to the six patients with lesion-based abscopal 95% CI (0.3 to 6.8), p=0.034) (table 2). response who did not achieve the criteria for RECIST- based abscopal response (p=0.89) (figure 2C). For irradi- Treatment ated lesions, diameter changes according to RECIST were Nearly all patients underwent hypofractionated RT (n=91 not associated with OS (p=0.72) (figures 2D–E and 3). of 94; 96.9%). Six patients had two irradiated metastases CD8 radiomics score and association with patient outcomes (6.4%), and one had three irradiated metastases (1.1%). At baseline, median value of the CD8 radiomics score Median number of fractions and dose by fraction given was not significantly different between the two groups of were three fractions (IQR (3–3)) of 8 Gy (IQR (6–12)). lesions (irradiated and non-irradiated) (1.75, IQR (1.55– Median number of IO therapy cycles given was 4 (IQR 1.97) vs 1.74, IQR (1.54–1.97), respectively, difference in (3–10)). Median time between the start of IO therapy and location=0.005, 95% CI (−0.03 to 0.04), p=0.76). Radio- the start of RT was 21 days (IQR (8–24)). Sixteen patients mics score of CD8 T-cells was not correlated with lesion started RT before IO therapy (median interval: −5.5, IQR size (Spearman’s rho=−0.005, p=0.93). (−7–−1.75)). Median follow-up was 14.8 months (IQR (8.4–20.8)). Median time between baseline and follow-up Lesion response CT was 2.8 months (IQR (2.0–3.4)). Association between the CD8 radiomics score of a lesion at baseline and its response according to RECIST was http://jitc.bmj.com/ Patient outcomes description significant (p=0.0066) with responding lesion having Lesion response a higher CD8 radiomics score (median (IQR)=1.95 The median per cent change in lesion diameter was (1.64–1.98) and 1.72 (1.53–1.97) for stable or progressive −13.0% (IQR (−28.9–+10.1)) for irradiated metastases lesions (area under the receiving operating character- versus +11.2% (IQR (−11.0–+40.5)) for non-irradiated istic curve (AUC)=0.63, 95% CI (0.56 to 0.71), p=0.0020) metastases (p<0.0001), with respectively 16% progressive (figure 3A–B). on September 25, 2021 by guest. Protected copyright. lesions (n=16 of 100 lesions) and 23% responding lesions (n=23 of 100) for irradiated metastases and 38% progres- Patient response sive lesions (n=72 of 189) and 17% responding lesions Several metrices were evaluated to decipher whether (n=32 of 189) for non-irradiated metastases. heterogeneity of the CD8 score at baseline was associated with clinical outcomes (figure 3C–D and online supple- Patient response mentary table S1). Regarding patient response at the first follow-up CT, 6 Patients with uniform progression had lower CD8 complete responses, 22 partial responses, 24 SD and 42 radiomics scores at baseline (based on the minimal value progressive disease were observed according to RECIST across their lesions) than those with uniform response V.1.1, yielding an objective RR of 29.9% (28 of 94 (p=0.012) and mixed response (p=0.0073), but patients patients). More precisely, 24.5% (n=23) showed uniform with either mixed response or uniform progression response, 38.3% (n=36) showed uniform progression, had more heterogeneous CD8 radiomics scores across 12.8% (n=12) showed mixed response and 24.5% (n=23) their lesions (higher entropy of the distribution of CD8 presented only stable lesions. T-cells radiomics score) than patients with SD or uniform Median OS was 25.2 months (95% CI (17.3–NA)) response (online supplementary figure S4). and median PFS was 4.7 months (95% CI (3.5–6.7)) Entropy of the distribution of the CD8 radiomics scores, (online supplementary figure S2). Patients with uniform was associated with PFS, patients with higher value than
4 Sun R, et al. J Immunother Cancer 2020;8:e001429. doi:10.1136/jitc-2020-001429 Open access J Immunother Cancer: first published as 10.1136/jitc-2020-001429 on 13 November 2020. Downloaded from Continued 0.179 0.089 0.012 0.583 0 (0.0) 0 (0.0) <0.001 <0.001 <0.001 11 (27.5) 16 (40.0) 13 (32.5) 21 (52.5) 19 (47.5) 40 (100.0) P value NA 5 (12.5) 0 (0.0) 0 (0.0) 0 (0.0) 0 (0.0) 0 (0.0) 2 (5.0) 7 (17.5) 0 (0.0) 0 (0.0) 3 (18.8) 2 (12.5) 3 (18.8) 40 10 (25.0) 30 (75.0) 35 (87.5) 25 (62.5) 13 (32.5) 33 (82.5) 13 (81.2) 14 (87.5) 13 (81.2) 0.00 (0.00 to 1.00) 3.00 (2.00 to 3.00) 1.00 (1.00 to 1.00) 2.00 (1.00 to 2.00) 3.00 (0.00 to 12.00) 5.50 (4.00 to 16.00) Ghent 2 (12.5) 0 (0.0) 0 (0.0) 0 (0.0) 0 (0.0) 5 (31.2) 0 (0.0) 0 (0.0) 0 (0.0) 3 (18.8) 0 (0.0) 0 (0.0) 0 (0.0) 0 (0.0) 16 14 (87.5) 16 (100.0) 11 (68.8) 13 (81.2) 24 (63.2) 14 (36.8) 38 (100.0) 38 (100.0) Erlangen 1 (2.6) 0 (0.0) 0 (0.0) 3 (7.9) 0 (0.0) 0 (0.0) 3 (3.2) 38 14 (36.8) 24 (63.2) 18 (47.4) 19 (50.0) 21 (55.3) 14 (36.8) 20 (52.6) 18 (47.4) 24 (25.5) 25 (26.6) 29 (30.9) 16 (17.0) 35 (37.2) 59 (62.8) 91 (96.8) 4.00 (3.00 to 10.00) 4.00 (3.00 to 6.00) GR http://jitc.bmj.com/ Overall 94 60.38 (50.68 to 69.04)26 (27.7) 57.38 (48.75 to 67.77)68 (72.3) 64.04 (55.84 to 69.12)53 (56.4) 24 (25.5) 62.80 (50.28 to 71.36)1 (1.1) 16 (17.0) 0.320 5 (5.3) 39 (41.5) 14 (14.9) 13 (13.8) 1.00 (0.00 to 1.00)85.50 (61.75 to 103.75) 1.00 (1.00 to 2.00) 63.00 (44.25 to 89.75)3.00 (2.00 to 4.00) 128.00 (106.25 to 141.00) NA (NA, NA) 1.00 (1.00 to 1.00) 83.50 (69.75 to 102.00) 3.00 (2.00 to 4.00)2.00 (1.00 to 3.00) 1.00 (1.00 to 1.00) <0.001 3.00 (3.00 to 4.25) 56 (59.6) 2.00 (1.00 to 3.00) 1.00 (1.00 to 2.00) 38 (40.4) 2.50 (1.00 to 3.00) on September 25, 2021 by guest. Protected copyright. 1 2 1 Level F M NA CRC adenocarcinoma 21 (22.3) HNSCC Melanoma NSCLC TCC Unknown primary 2 (2.1) < or =2 > or =3 Atezolizumab Ipilimumab Nivolumab Pembro < or =14 days > day 14 irradiated Patient characteristics IQR ))
Table 1 Table lesions at baseline (median( n Age (median(IQR )) Sex (% ) Performance status (% ) 0 Histology (% ) Number of prior lines therapy (median(IQR )) CT interval (median(IQR )) N of VOI at baseline (median(IQR )) N of VOI irradiated lesions at baseline (median(IQR )) N of VOI non- N of organs involved (%) IO therapy (%) N of IO cycles (median(IQR )) and IO Time between RT therapy (% ) (% )Hypofractionated RT 0
Sun R, et al. J Immunother Cancer 2020;8:e001429. doi:10.1136/jitc-2020-001429 5 Open access J Immunother Cancer: first published as 10.1136/jitc-2020-001429 on 13 November 2020. Downloaded from Continued 0 (0.0) 4 (10.0) 0 (0.0) 7 (17.5) 0 (0.0) 6 (15.0) 0 (0.0) 0 (0.0) 0 (0.0) 5 (12.5) 4 (10.0) 2 (5.0) 0 (0.0) 0 (0.0) 40 (100.0) 36 (90.0) 33 (82.5) 34 (85.0) 29 (72.5) 11 (27.5) 15 (37.5) 25 (62.5) 31 (77.5) 13 (32.5) 25 (62.5) 10 (25.0) P value 7 (43.8) 9 (56.2) 5 (31.2) 2 (12.5) 9 (56.2) 1 (6.2) 6 (37.5) 9 (56.2) 2 (12.5) 5 (31.2) 9 (56.2) 6 (37.5) 2 (12.5) 8 (50.0) 1 (6.2) 4 (25.0) 9 (56.2) 7 (43.8) 0 (0.0) 0 (0.0) 5 (31.2) 1 (6.2) 0 (0.0) 0 (0.0) 11 (68.8) 10 (62.5) 8.75 (5.00–17.50) 1.00 (0.00–10.00) 30.00 (12.50–80.00) 1.00 (0.00–10.00) Ghent 0 (0.0) 0 (0.0) 8 (21.1) 0 (0.0) 9 (23.7) 0 (0.0) 0 (0.0) 5 (13.2) 1 (2.6) 0 (0.0) 6 (15.8) 3 (7.9) 1 (2.6) 38 (100.0) 27 (71.1) 11 (28.9) 30 (78.9) 29 (76.3) 21 (55.3) 17 (44.7) 32 (84.2) 18 (47.4) 20 (52.6) 10 (26.3) 18 (47.4) 16 (42.1) Erlangen 7 (7.4) 9 (9.6) 9 (9.6) 9 (9.6) 8 (8.5) 6 (6.4) 4 (4.3) 7 (7.4) 3 (3.2) 87 (92.6) 68 (72.3) 17 (18.1) 64 (68.1) 21 (22.3) 65 (69.1) 20 (21.3) 56 (59.6) 30 (31.9) 58 (61.7) 30 (31.9) 58 (61.7) 32 (34.0) 23 (24.5) 25 (26.6) 36 (38.3) 26 (27.7) 3.00 (3.00 to 3.00)8.00 (6.00 to 12.00) 3.00 (3.00 to 3.00) 13.00 (5.00 to 15.00) 12.00 (10.75 to 15.00)5.00 (0.00–33.75) 3.50 (3.00 to 6.00) 3.00 (3.00 to 3.00) 5.00 (0.00–12.50) NA (NA, NA) 8.00 (8.00 to 8.00) NA (NA, NA) 36.00 (36.00 to 69.00) 75.00 (18.75 to 93.75) 51.55 (41.25 to 70.00) 36.00 (36.00 to 36.00) GR 451.50 (256.50 to 633.75) 360.50 (243.00 to 624.75) 277.50 (187.50 to 394.25) 540.00 (376.00–817.75) http://jitc.bmj.com/ G/L
mg/L
G/L
G/L
mg/L
G/L
Overall < or =7.5 >7.5 NA > or =1 <1 NA >6 NA < or =250 >250 NA > or =35 <35 NA < or =10 >10 NA 0 1 2 3 NA Low on September 25, 2021 by guest. Protected copyright. ) < or =6 IQR )) 0 1 Level IQR )) lymphocyte ratio (% to- Continued
up (median( IQR )) L1 (tumor cells) (median( L1 (immune cells) (median( Stereotactic RT (% ) RT Stereotactic N of fractions RT: (median(IQR )) dose per fraction RT: (median(IQR )) total dose in EQD2 RT: (median(IQR )) Follow- PD- PD- count (% ) Baseline absolute neutrophil Baseline absolute lymphocyte count (% ) Baseline neutrophil- Baseline LDH (% ) Baseline albumin (% ) Baseline CRP (% ) (% ) Baseline RMH score (% ) Baseline GR immune score Table 1 Table
6 Sun R, et al. J Immunother Cancer 2020;8:e001429. doi:10.1136/jitc-2020-001429 Open access J Immunother Cancer: first published as 10.1136/jitc-2020-001429 on 13 November 2020. Downloaded from
the median having poorer PFS (HR=1.67, 95% CI (1.02 to 2.75), p=0.040, FDR=0.080), and remained significantly associated with OS (HR=2.08, 95% CI (1.09 to 3.96), p=0.023, FDR=0.17). oncology; Minimal value, maximal value, mean and SD of the lesion scores were not significantly associated with PFS or 5 (12.5) 25 (62.5) OS. However, a statistical trend was identified between PFS
P value and the minimal value of the CD8 radiomics score across all lesions (p=0.07), and significance was reached when looking only at patients with extreme values (HR=0.34,
e; RT, radiotherapy; TCC, e; RT, 95% CI (0.17 to 0.68), p=0.0021) (online supplementary table S1). CD8 radiomics scores on the follow-up CT were not significantly different from baseline CT. 3 (18.8) 13 (81.2) Irradiated and non-irradiated lesions subgroup analysis Ghent On a lesion level, for both irradiated and non-irradiated lesions, the radiomics scores of responding lesions were higher than those of progressive lesions (AUC=0.75, 95% CI (0.58 to 0.92), p=0.0090 and AUC=0.75, 95% CI (0.51 to 0.74), p=0.040, respectively). There was no differ- ence in tumor size at baseline between responding and
1 (2.6) progressive lesions (online supplementary figure S5). 21 (55.3) On a patient level, minimum value of the CD8 radio- ligand 1; RMH, Royal Marsden Hospital scor
Erlangen mics score across non-irradiated lesions was significantly associated with PFS (HR=0.37, 95% CI (0.18 to 0.76), p=0.0067, FDR=0.022). For the irradiated lesions, this value was not significantly associated with PFS (HR=0.62, 95% CI (0.38 to 1.02), p=0.059, FDR=0.10). When looking
ogrammed death- at the extreme values, the minimum values of both non- irradiated and irradiated lesions were associated with PFS L1, pr 29 (30.9) 39 (41.5) (p=0.0011 and p=0.0067, respectively).
GR Regarding out-of-field response, heterogeneous CD8 radiomics scores based on entropy of the distribution http://jitc.bmj.com/ Gy fraction; GR, Gustave Roussy ; HNSCC, squamous cell cancers of the head and neck; IO, immuno- were associated with RECIST-based abscopal response (AUC=0.70, 95% CI (0.56 to 0.84), p=0.014).
Clinical and biological predictors of outcomes and multivariate analysis small cell lung cancer; PD- Univariate analysis of OS and PFS are summarized in Overall High NA
supplemental data (online supplementary table S2, S3). on September 25, 2021 by guest. Protected copyright. A performance status (PS) >0, an interval between IO therapy and RT start >14 days, high level of lactate dehy- drogenase (>250 UI/L), high level of C reactive protein (CRP) (>10 mg/L) and high entropy of the CD8 T-cells radiomics scores distribution were significantly associated with both poorer OS and PFS. In the multivariate anal- ysis of OS including these factors, only CRP (HR=3.34; Level 95% CI (1.48 to 7.52), p=0.0036) and the entropy of the CD8 T-cells radiomics scores distribution remained signif- icant (HR=2.64; 95% CI (1.21 to 5.75), p=0.015) (online
ogenase; NA, not available; NSCLC, non- supplementary table S4). For PFS, only the PS remained significant in multivariate analysis (HR=4.6; 95% CI (2.28 to 9.28), p<0.0001) (online supplementary table S4). Continued
Center of inclusion was not a significant confounding factor when included in the multivariate analyses (online supplementary table S5). In this cohort, we did not Transitional cell carcinoma; VOI, volume of interest. cell carcinoma; Transitional Table 1 Table EQD2, equivalent dose in 2 protein; C reactive cancer; CRP, CRC, colorectal LDH, lactate dehydr identify a prognostic role of PD-L1 either for PFS or OS
Sun R, et al. J Immunother Cancer 2020;8:e001429. doi:10.1136/jitc-2020-001429 7 Open access J Immunother Cancer: first published as 10.1136/jitc-2020-001429 on 13 November 2020. Downloaded from
Table 2 Characteristics of lesions analyzed at baseline Level Irradiated Non-irradiated P value n 100 189 Location (%) Node 38 (38.0) 64 (33.9) 0.015 Brain 1 (1.0) 0 (0.0) Liver 14 (14.0) 42 (22.2) Head and neck 1 (1.0) 0 (0.0) Bone 4 (4.0) 0 (0.0) Other 17 (17.0) 42 (22.2) Lung 25 (25.0) 39 (20.6) Adrenal gland 0 (0.0) 2 (1.1) Lesion size (median (IQR)) 26.92 (17.55 to 37.71) 21.54 (15.23 to 31.89) 0.034 Lesion volume (median (IQR)) 4.84 (1.26 to 15.74) 2.83 (0.99 to 9.26) 0.093 CD8 radiomics score (median (IQR)) 1.75 (1.55 to 1.97) 1.74 (1.54 to 1.97) 0.756
(online supplementary table S2, S3). Royal Marsden lesions with the highest radiomics scores would be more Hospital score and Gustave Roussy Immune Score were likely to present out-of- field response to IO therapy. associated with PFS, but not with OS.14 26–29 Indeed, association with PFS and OS was even more important when considering patients with extreme values (although not significant for OS in this study). Entropy of DISCUSSION the distribution of the radiomics scores of CD8 T-cells was IORT is becoming widely used to potentiate immune associated with both PFS and OS and could discriminate response. Imaging biomarkers would be very valuable patients who will present uniform response from those as they could assess the ensemble of the disease extent with uniform PD and mixed response. A high entropy and its spatial heterogeneity, and may help in the choice indicated heterogeneous scores of CD8 infiltration and of the target lesion(s) to irradiate in order to optimize was associated with poorer outcomes. This index high- synergistic effects.10 30 Some authors hypothesized that lighted the impact of tumor heterogeneity on the prog- interlesion heterogeneity of bulky disease may limit the nosis of the patient and the importance of assessing the probability of RT-induced systemic antitumor immune whole disease instead of only one lesion. response, especially in single-site abscopal approach.9 These results confirm the potential of radiomics to Moreover, some studies suggested that the location of http://jitc.bmj.com/ assess spatial heterogeneity as a non-invasive ‘multiple irradiation may impact how the immune response is elic- virtual biopsies’.14 Korpics et al recently presented similar ited, but there is still no evidence regarding the optimal 31 preliminary results in a population of patients treated strategy. with IORT, adding further external validation, although In this study, we demonstrated that the predictive value 24 of a radiomics score of CD8 T-cells previously validated in they only evaluated irradiated lesions. They analyzed 68 patients and 139 irradiated tumors and showed associa- a cohort of patients treated with IO monotherapy could on September 25, 2021 by guest. Protected copyright. still be informative in a context of radio-immunotherapy tions between the CD8 radiomics score and local tumor combination.13 Indeed, lesions with high radiomics response (OR=10.21, 95% CI (1.76 to 59.17); p=0.010), score at baseline were associated with decrease in the OS (HR=0.39, 95% CI (0.20 to 0.75); p=0.005) and PFS lesion size at the follow-up CT, irrespective of irradiation. (HR=0.47, 95% CI (0.26 to 0.85); p=0.013), using a 25% Moreover, our results suggest that this tool could also be percentile cut-off for dichotomization of patients. used to assess tumor heterogeneity, which was associated To the best of our knowledge, this previously published with patient outcome. On a per-patient basis, the lowest radiomics signature of CD8 T-cells predicting response to radiomics score of the patient’s non-irradiated lesions IO is the first to be validated by several subsequent inde- 13 24 was associated with PFS. This index was concordant pendent studies. We acknowledge however that this with the observation that outcomes of non-irradiated study had some limitations. The magnitude of the asso- lesions (vs irradiated lesion) were more predictive of the ciation between the CD8 radiomics score and the lesion patient outcome (figures 2D–E and online supplemen- response, although significant, was limited with over- tary figure S4). Therefore, this radiomics score of CD8 laps between the different groups of response. However, T-cells could be used to identify the lesions with the lowest considering that the radiomic signature has not been CD8 tumor infiltration. Such lesions could benefit from retrained in this cohort, it was interesting to note that local treatment since irradiation was associated with more results seemed concordant with those published in the responding lesions. On the contrary, the non-irradiated original study (radiomics score between patients with
8 Sun R, et al. J Immunother Cancer 2020;8:e001429. doi:10.1136/jitc-2020-001429 Open access J Immunother Cancer: first published as 10.1136/jitc-2020-001429 on 13 November 2020. Downloaded from http://jitc.bmj.com/ on September 25, 2021 by guest. Protected copyright.
Figure 2 (A) Response kinetics curve depicting individual lesion responses (red dots corresponding to irradiated lesions, blue dots corresponding to non-irradiated lesions) on a patient-to-patient basis. (B) Kaplan-Meier curves of overall survival according to patients’ patterns of response. (C) Kaplan-Meier curves of overall survival according to out-of- field esponser (OOF). Out-of- field esponser which fulfilled RECIST V.1.1 criteria (30% decrease in the sum of the aggregate diameters of the non-irradiated lesions) showed higher overall survival (OS) than the rest of the population. (D, E) Kaplan-Meier curves of overall survival according to aggregate diameter changes of irradiated lesions (D) or non-irradiated lesions (E).
Sun R, et al. J Immunother Cancer 2020;8:e001429. doi:10.1136/jitc-2020-001429 9 Open access J Immunother Cancer: first published as 10.1136/jitc-2020-001429 on 13 November 2020. Downloaded from http://jitc.bmj.com/ on September 25, 2021 by guest. Protected copyright.
Figure 3A Lesion response according to baseline radiomics score. Responding lesions corresponded to lesions with a decrease in tumor size >30% (partial response and complete response). (B) Scatter plot of changes in tumor size (%) according to baseline radiomics score. (C, D) Associations between distribution metrices of the radiomics score of CD8 T-cells and clinical outcomes (progression-free survival (PFS), overall survival (OS)).
Complete Response or Partial Response (CRPR) and interpreted with caution and should be only considered Stable Disease or Progressive Disease (SDPD) at 3 months as exploratory. Overall, validation of these data still needs ‘Wilcoxon’s difference in location 0.12, 95% CI 9.9×10−⁶ to be done in prospective and, if possible, randomized to 2.3; p=0.049’, corresponding to an AUC of 0.62 in trials with homogeneous patients to refine and improve the immunotherapy cohort).13 Therefore, these results the performance of the signature in specific indications. comfort further the reliability of the validated signa- Moreover, the association between the IORT protocol ture. Given the relatively low number of patients and the and the primitive tumor type inherent to the different number of lesions evaluated per patient, we acknowledge clinical trials may have brought potential confounding that the analysis of the spatial heterogeneity should be factors limiting the analysis of the predictive factors. Still,
10 Sun R, et al. J Immunother Cancer 2020;8:e001429. doi:10.1136/jitc-2020-001429 Open access J Immunother Cancer: first published as 10.1136/jitc-2020-001429 on 13 November 2020. Downloaded from this study addressed several pending questions regarding content: NS, MH, FP, AL, NP, BF, J-CS, CM, CF, RF, PO, UG. Statistical analysis: RS, IORT. Tumor heterogeneity has shown to be important MV, EB. Obtained funding: RS, CF, FP, CM, ED. Administrative, technical or material support: BF, CM, RF, PO, J-CS, UG, ED, CR, AC, Estienne, Alvarez-Andres, Niyoteka. and correlates to certain extent with outcomes. However, Study supervision: ED. the ‘abscopal’ out-of-field response should not be based Funding Fondation pour la Recherche Médicale (grant DIC20161236437), only on the response of one non-irradiated lesion, as Fondation ARC pour la recherche contre le cancer (grant SIGN'IT20181007805), 32 frequently reported. OS of patients who do not fulfill SIRIC-SOCRATE 2.0 (grant INCa-DGOS-INSERM_12551) and Amazon AWS grant out-of- field RECIST criteria may not be different from provided funding to INSERM U1030 Radiomics team—Gustave Roussy. Fondation patients without out-of- field response. Rather, we should ARC pour la recherche contre le cancer and Paris-Saclay University provided support to RS with a grant for international mobility. RS has received the support of stick with RECIST V.1.1. criteria which recommend not INSERM, in partnership with Fondation Bettencourt Schueller, within the program of to use irradiated lesions as RECIST-target lesions to assess CCA-INSERM- BETTENCOURT. treatment response, and analyze several non-irradiated Competing interests RS received travel and accommodation expenses from RECIST-target lesions to have information about the AstraZeneca. NS reports travel grants from Merck Sharpe & Dohme, Astellas, Bayer whole disease and the response heterogeneity.19 33 34 and Bristol-Myers Squibb. MH declares consulting fees, research grants and travel expenses from Merck Serono, MSD, Novartis, BMS and Teva. PO reports research Regarding RT modalities, an interval between RT and grants from Merck Sharpe & Dohme, Astellas and Janssen; travel grants from IO therapy of <2 weeks was associated with better PFS Ipsen and Ferring Pharmaceuticals and honorarium from Ferring and Bayer. CR and OS, although not significant in multivariate analysis. received research grants from Therapanacea and Elekta. CF is a full-time employee Interestingly, some variables such as liver metastasis, or of AstraZeneca since September 2017. NP reports grants from GE, Biogen and consulting fees from AstraZeneca and Ipsen. J-CS reports consultancy fees from scores like the Royal Marsden Hospital score and the AstraZeneca, Astex, Clovis, GSK, GamaMabs, Lilly, MSD, Mission Therapeutics, Gustave Roussy Immune Score known to be prognostic Merus, Pfizer, PharmaMar, Pierre Fabre, Roche-Genentech, Sanofi, Servier, factors with IO therapy,14 26–29 were only associated with Symphogen, Takeda. J-CS has been Full time employee for AstraZeneca (September PFS and not with OS in this cohort of patients treated with 2017–December 2019) and is a shareholder of AstraZeneca, Gritstone. CF received travel, accommodations, expenses from Amgen; Genentech; GlaxoSmithKline; IORT. This may reflect potential long-lasting RT-induced MedImmune. CM is an advisory board member, speaker, investigator at Amgen, immune response for which new biomarkers specifically Astellas, AstraZeneca, Bayer, Celgene, Genentech, Ipsen, Janssen, Eli Lilly, Novartis, validated for IORT strategies should be identified. Pfizer, Roche, Sanofi, Orion. USG received support for presentation activities for Dr Sennewald Medizintechnik GmbH, has received support for investigator initiated clinical studies (IITs) from MSD and AstraZeneca and contributed at Advisory Boards Meetings of AstraZeneca and Bristol-Myers Squibb. RF received grants/research CONCLUSION support from MSD; AstraZeneca; Merck (Germany/Darmstadt); Novocure; honoraria/ consultation fees: MSD; AstraZeneca; Merck (Germany/Darmstadt); Novocure; Our findings confirm the predictive value of the previ- Fresenius and reports participation in a company sponsered speakers bureau: ously published CD8 radiomics score in patients with AstraZeneca; MSD; Merck (Germany/Darmstadt); Sennewald; Novocure; Brainlab; solid tumors treated with IORT combinations.13 More- Roche. ED has declared consulting fees and support from Roche, BMS, Boehringer, over, the results suggest that global spatial reasoning on AstraZeneca, Lilly Amgen and Merck-Serono. this signature through a statistical analysis of the distri- Patient consent for publication Not required. bution of this score across patient’s lesions could be a Ethics approval The study protocol was approved by the Medical Ethics promising method to assess tumor heterogeneity and to Committee of each center. Patients provided informed consent in accordance with predict clinical outcomes. Overall, this study asserts the their respective trial protocols. http://jitc.bmj.com/ potential of radiomics to help clinicians characterize the Provenance and peer review Not commissioned; externally peer reviewed. whole disease and to move toward in precision medicine. Data availability statement All data relevant to the study are included in the article or uploaded as supplementary information. Author affiliations Supplemental material This content has been supplied by the author(s). It 1Department of Radiation Oncology, Gustave Roussy, Villejuif, Île-de- France, France has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have 2 Institut Gustave Roussy, Inserm, Radiothérapie Moléculaire et Innovation been peer-reviewed. Any opinions or recommendations discussed are solely on September 25, 2021 by guest. Protected copyright. Thérapeutique, Paris-Saclay University, Villejuif, Île-de-F rance, France those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability 3Paris- Saclay University Faculty of Medicine, Le Kremlin-Bicetre, Île-de-F rance, and responsibility arising from any reliance placed on the content. Where the France content includes any translated material, BMJ does not warrant the accuracy and 4Department of Radiation Oncology, University Hospital Ghent, Gent, Oost- reliability of the translations (including but not limited to local regulations, clinical Vlaanderen, Belgium guidelines, terminology, drug names and drug dosages), and is not responsible 5Department of Radiation Oncology, Friedrich-Alexander-Universität Erlangen- for any error and/or omissions arising from translation and adaptation or otherwise. Nürnberg, Erlangen, Germany 6Department of Radiation Oncology, Fondazione IRCCS Policlinico San Matteo, Open access This is an open access article distributed in accordance with the Pavia, Lombardia, Italy Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits 7TheraPanacea, Paris, France others to copy, redistribute, remix, transform and build upon this work for any 8Drug Development Department, Gustave Roussy, Villejuif, Île-de-F rance, France purpose, provided the original work is properly cited, a link to the licence is given, 9Departement of Medicine, Gustave Roussy, Villejuif, Île-de-F rance, France and indication of whether changes were made. See https://creativecommons.org/ 10CentraleSupélec, Gif-sur- Yvette, Île-de- France, France licenses/by/ 4.0/. ORCID iDs Twitter Roger Sun @rgrsun Roger Sun http://orcid. org/0000- 0001-9866- 6449 Contributors RS, MV and ED had full access to all of the data in the study and take Markus Hecht http://orcid. org/0000- 0003-2082- 216X responsibility for the integrity of the data and the accuracy of the data analysis. Charlotte Robert http://orcid. org/0000- 0001-8838- 1343 Study concept and design: RS, NP, MV, ED. Acquisition, analysis or interpretation Celine Boutros http://orcid. org/0000- 0001-5659- 8791 of data: RS, NS, MH, FP, AL, AR, MM, ER, GL, JD-L, SB, RB, CB. Drafting of the Benjamin Frey http://orcid. org/0000- 0001-6743- 3351 manuscript: RS, ED. Critical revision of the manuscript for important intellectual Udo Gaipl http://orcid. org/0000- 0001-6375- 5476
Sun R, et al. J Immunother Cancer 2020;8:e001429. doi:10.1136/jitc-2020-001429 11 Open access J Immunother Cancer: first published as 10.1136/jitc-2020-001429 on 13 November 2020. Downloaded from
REFERENCES 19 Sundahl N, Seremet T, Van Dorpe J, et al. Phase 2 trial of nivolumab 1 Robert C, Schachter J, Long GV, et al. Pembrolizumab versus combined with stereotactic body radiation therapy in patients with ipilimumab in advanced melanoma. N Engl J Med 2015;372:2521–32. metastatic or locally advanced inoperable melanoma. Int J Radiat 2 Reck M, Rodríguez-Abr eu D, Robinson AG, et al. Pembrolizumab Oncol Biol Phys 2019;104:828–35. versus chemotherapy for PD-L1-positive non-small-cell lung cancer. 20 Sundahl N, De Wolf K, Kruse V, et al. Phase 1 Dose Escalation Trial N Engl J Med 2016;375:1823–33. of Ipilimumab and Stereotactic Body Radiation Therapy in Metastatic 3 Chen DS, Mellman I. Elements of cancer immunity and the cancer- Melanoma. Int J Radiat Oncol Biol Phys 2018;100:906–15. immune set point. Nature 2017;541:321–30. 21 Sundahl N, Vandekerkhove G, Decaestecker K, et al. Randomized 4 Weichselbaum RR, Liang H, Deng L, et al. Radiotherapy and phase 1 trial of pembrolizumab with sequential versus concomitant immunotherapy: a beneficial liaison? Nat Rev Clin Oncol stereotactic body radiotherapy in metastatic urothelial carcinoma. 2017;14:365–79. Eur Urol 2019;75:707–11. 5 Demaria S, Ng B, Devitt ML, et al. Ionizing radiation inhibition of 22 Nioche C, Orlhac F, Boughdad S, et al. A freeware for tumor distant untreated tumors (abscopal effect) is immune mediated. Int J heterogeneity characterization in PET, SPECT, CT, MRI and us to Radiat Oncol Biol Phys 2004;58:862–70. accelerate advances in radiomics. J Nucl Med 2017;58:1316. 6 Twyman-Saint Victor C, Rech AJ, Maity A, et al. Radiation and dual 23 Leijenaar RTH, Nalbantov G, Carvalho S, et al. The effect of SUV checkpoint blockade activate non-redundant immune mechanisms in discretization in quantitative FDG-PET Radiomics: the need for cancer. Nature 2015;520:373–7. standardized methodology in tumor texture analysis. Sci Rep 7 Frey B, Rückert M, Deloch L, et al. Immunomodulation by ionizing 2015;5:11075. radiation-impact for design of radio-immunotherapies and for 24 Korpics MC, Polley M-Y, Bhave SR, et al. A Validated T Cell treatment of inflammatory diseases. Immunol Rev 2017;280:231–48. Radiomics Score is Associated with Clinical Outcomes Following 8 Rückert M, Deloch L, Fietkau R, et al. Immune modulatory effects Multi-site SBRT and Pembrolizumab [Internet]. International Journal of radiotherapy as basis for well-reasoned radioimmunotherapies. of Radiation Oncology*Biology*Physics 2020. Strahlenther Onkol 2018;194:509–19. 25 R Core Team. R: A Language and Environment for Statistical 9 Brooks ED, Chang JY. Time to abandon single-site irradiation for Computing [Internet]. Vienna, Austria: R Foundation for Statistical inducing abscopal effects. Nat Rev Clin Oncol 2019;16:123–35. Computing, 2017. Available: https://www.R-project.org 10 Deutsch E, Chargari C, Galluzzi L, et al. Optimising efficacy and 26 Minami S, Ihara S, Ikuta S, et al. Gustave Roussy immune score and reducing toxicity of anticancer radioimmunotherapy. Lancet Oncol Royal Marsden Hospital prognostic score are biomarkers of Immune- 2019;20:e452–63. Checkpoint inhibitor for non-small cell lung cancer. World J Oncol 11 Limkin EJ, Sun R, Dercle L, et al. Promises and challenges for the 2019;10:90–100. implementation of computational medical imaging (radiomics) in 27 Wagner M, Hess KR, Hong DS, et al. Validation of the Royal Marsden oncology. Ann Oncol 2017;28:1191–206. Hospital (RMH) prognostic score in 100 patients with advanced 12 Gerlinger M, Rowan AJ, Horswell S, et al. Intratumor heterogeneity sarcoma enrolled in early phase clinical trials at a major cancer and branched evolution revealed by multiregion sequencing. N Engl center. Journal of Clinical Oncology 2015;33:10558. J Med 2012;366:883–92. 28 Bigot F, Castanon E, Baldini C, et al. Prospective validation of a 13 Sun R, Limkin EJ, Vakalopoulou M, et al. A radiomics approach to prognostic score for patients in immunotherapy phase I trials: assess tumour-infiltrating CD8 cells and esponser to anti-PD-1 or the Gustave Roussy immune score (GRIm-Scor e). Eur J Cancer anti-PD-L1 immunotherapy: an imaging biomarker, retrospective 2017;84:212–8. multicohort study. Lancet Oncol 2018;19:1180–91. 29 Tumeh PC, Hellmann MD, Hamid O, et al. Liver metastasis and 14 Trebeschi S, Drago SG, Birkbak NJ, et al. Predicting response to treatment outcome with anti-PD-1 monoclonal antibody in patients cancer immunotherapy using noninvasive radiomic biomarkers. Ann with melanoma and NSCLC. Cancer Immunol Res 2017;5:417–24. Oncol 2019;30:998–1004. 30 Spaas M, Lievens Y. Is the combination of immunotherapy and 15 Khorrami M, Prasanna P, Gupta A, et al. Changes in CT radiomic radiotherapy in non-small cell lung cancer a feasible and effective features associated with lymphocyte distribution predict overall approach? Front Med 2019;6:244. survival and response to immunotherapy in non-small cell lung 31 Tang C, Welsh JW, de Groot P, et al. Ipilimumab with stereotactic cancer. Cancer Immunol Res 2020;8:108–19. ablative radiation therapy: phase I results and immunologic 16 Bhatia A, Birger M, Veeraraghavan H, et al. Mri radiomic features are correlates from peripheral T cells. Clin Cancer Res 2017;23:1388–96. associated with survival in melanoma brain metastases treated with 32 Golden EB, Chhabra A, Chachoua A, et al. Local radiotherapy and immune checkpoint inhibitors. Neuro Oncol 2019;21:1578–86. granulocyte-macrophage colony-stimulating factor to generate
17 Tunali I, Gray JE, Qi J, et al. Novel clinical and radiomic predictors abscopal responses in patients with metastatic solid tumours: a http://jitc.bmj.com/ of rapid disease progression phenotypes among lung cancer proof-of-principle trial. Lancet Oncol 2015;16:795–803. patients treated with immunotherapy: an early report. Lung Cancer 33 Eisenhauer EA, Therasse P, Bogaerts J, et al. New response 2019;129:75–9. evaluation criteria in solid tumours: revised RECIST guideline (version 18 Alessandrino F, Gujrathi R, Nassar AH, et al. Predictive role of 1.1). Eur J Cancer 2009;45:228–47. computed tomography texture analysis in patients with metastatic 34 Luke JJ, Lemons JM, Karrison TG, et al. Safety and clinical activity urothelial cancer treated with programmed death-1 and programmed of pembrolizumab and multisite stereotactic body radiotherapy in Death-ligand 1 inhibitors. Eur Urol Oncol 2020;3:680–6. patients with advanced solid tumors. J Clin Oncol 2018;36:1611–8. on September 25, 2021 by guest. Protected copyright.
12 Sun R, et al. J Immunother Cancer 2020;8:e001429. doi:10.1136/jitc-2020-001429